CHAPTER SIXTEEN



GWO-WEI TORNG



A SPATIAL PERSPECTIVE OF THE INTERSTATE HIGHWAY SYSTEM'S 

EFFECTS ON SUBURBANIZATION



Introduction

        The role of transportation systems in fostering growth and 

affecting land use structures has been of great interest to academics, 

planners, investors, and politicians.  Peter O. Muller formulated his 

explanation of urban form in his article, "Transportation and Urban Form: 

Stages in the Spatial Evolution of the American Metropolis", by 

identifying four transportation eras (Figure 1).  Figure 1 clearly 

illustrates the concept of dependency of land use on transportation 

system changes.  Indeed, innovations in transportation technologies in 

the past have greatly influenced people's travel attitudes, travel 

behaviors, and then extended the boundaries of socio-economic activities, 

such as workplaces, residential areas, and shopping.  While examining 

Figure 1, it is important to distinguish the growth effects from the 

distribution effects.  In Figure 1, the distribution effects can be 

interpreted as the changing shapes of urban forms, while the growth 

effects are the net change of total areas within the urban boundaries.





Figure 1, Intraurban Transport Eras and Metropolitan Growth Patterns: (I) 

Walking-Horsecar Era, (II) Electric Streetcar Era, (III) Recreational 

Auto Era, (IV) Freeway Era.

Source: Peter O. Muller, Transportation and Urban Form: Stages in the 

Spatial Evolution of the American Metropolis.
        The effects of transportation infrastructure on shaping urbanized 

areas have drawn significant attention in the past.  The relationship 

between transportation and land use is interactive.  Empirical urban 

development research reveals that land use patterns influence 

transportation investment decisions, which may later affect further land 

use changes.  In this research, only the effects of highway/expressway 

systems on urbanized area growth will be explored.

        The primary intention in building interstate highways was for 

"interstate" transportation purposes, e.g., national defense and 

interstate commerce.  Nevertheless, several major metropolitan 

transportation studies, e.g., a study done by the Chicago Area 

Transportation Study (CATS) in 1955, suggested that one purpose in 

developing the transportation networks was to serve travel generated by 

projected land-use expansions (Teaford, 1986).  Demand-driven 

transportation planning schemes were very much a part of the thinking.  

Indeed, more than 30 years' experience tells us that those limited-access 

highways have very significant impacts at the local level, in addition to 

their impact on land use at the regional or even broader geographical 

scale.  Rapid urban sprawl in the past several decades is perceived to be 

directly or indirectly induced by the emergence of highway systems.  

Because interstate highways are relatively "lineal" featured (one 

dimensional) with respect to overall urban areas, their effects should be 

considered in terms of directions as well as densities.

        This research examines the land use impact of interstate highway 

investments from a geographical perspective.  It begins with a brief 

literature review which summarizes the objectives, motivation, 

methodology, and results of previous research on similar issues.  In the 

second section, major research questions are illustrated.  Next, the 

methodologies applied in this research, e.g., grid-cell based raster 

spatial analysis, are introduced followed by the development of models 

and discussions of modeling results.  Then, the transition of spatial 

effects is analyzed based on the derived modeling results.  In the final 

section, conclusions drawn from research findings are summarized and some 

policy implications derived from this study are discussed. 



Literature Review

        Muller (1986) pointed out that Americans were not urban dwellers 

by design.  Many people live for a dream of owning a private, single 

family home, near-to-nature, rural, with huge front and back yards.  As 

described by Schaeffer and Sclar (1975), "people have found a way to 

isolate themselves;... a way to privacy among their peer group;... They 

see nothing except what they are determined to see."  For a majority of 

urban dwellers, living within large cities was actually not a personal 

preference but a personal choice given no other viable choices.  Starting 

from 1945, after a long period of economic depression and war, the U.S. 

showed a great potential for economic revitalization.  The automobile was 

no longer a luxury good but a necessity for commuting, shopping, 

socializing, and running personal errands.  Meanwhile, America's birth 

rate rose, many young couples left crowded central city neighborhoods and 

moved to spacious suburbs where their kids could play in their own front 

or back yards. Two-car families became commonplace.  In many aspects, the 

suburbanization appeared to be an inevitable tendency (Teaford, 1986).  

The passage of the Interstate Highway Act of 1956 signaled the arrival of 

the suburbanization era.  The booming highway systems acted as catalyst 

which turned the historic metropolitan cities inside out and reshaped the 

traditional urbanized areas .  This suburbanization phenomenon can be 

well explained by the increase in people's mobility.  People can afford 

to live outside the central cities by traveling at a much higher travel 

speed on the newly constructed highways while not increasing travel 

time.  Figure 2 illustrates the concept of the dynamic relationships 

between population density and urban area expansion from a spatial 

perspective.  The resulting increase in mobility indeed reduced the 

relative attractiveness of old and crowded central cities.





Figure 2, Type of Urban Area Expansion: (I) Pure Growth (increased 

population are even distributed), (II) No Growth, Redistribution (no 

change in total population), and (III) Growth + Redistribution (net gain 

in population and population redistribution)



        Meanwhile, the gasoline cost in the U.S. was low enough to 

support the highway commute choice.  Anthony Down (1992) pointed out that 

the inflation adjusted price of unleaded gasoline actually went down by 

27.4 percent from 1975 to 1987.  The gasoline prices in other industrial 

countries in 1991 were at least 50 percent higher than those in the 

United States.  In some areas, like Italy and Japan, the gasoline prices 

were three times or higher compared to the prices in the U.S.  The 

relatively lower gasoline cost reduces the auto operating cost and thus 

indirectly encourages the urban decentralization.

        Nevertheless, it is not appropriate to attribute the only cause 

of suburbanization to the transportation systems.  Bradbury, et al. 

(1982) concluded that several possible causes of urban decentralization 

exist.  The results suggested that national economic conditions, 

technological change, and  federal policies were also responsible for the 

urban decline in the past decades.  

        There are some studies concerning highway impacts on land use 

changes.  In the early interstate highway era, the majority of research, 

such as Adkins (1959), and Mohring (1961), tended to justify the positive 

effects of interstate highways through economic reasons, such as increase 

of land value.  However, these studies failed to address the difference 

between growth and distribution effects.  That is, whether the land value 

increases near the highway were simply a distributional shift or a net 

gain within the region as a whole was an issue.  Both researches used the 

highway length as a indicator rather than using the accessibility 

measures of the highway interchange.  It was recognized later that the 

highway interchange is a more meaningful and appropriate research unit 

when exploring  highway systems' impact on land use.  Finally, both 

researches were conducted in the late 1950s and early 1960s which might 

be too early to capture "time lag" effects of the highway investment.  

Thus an objective evaluation of the effects of these newly constructed 

highways was virtually impossible.

        A more recent and comprehensive highway impact studies was 

conducted by Payne-Maxie Consultants for U.S. Department of 

Transportation (DOT) and U.S. Department of Housing and Urban Development 

(HUD) in 1980.  This study was designed specifically to explore the 

impact of beltways (circumferential limited-access highways) instead of 

overall interstate highways.  It considered influential factors, such as 

city ages and highway location.  In addition, this study applied the 

concept of interchange density using it as a measurement of highway 

accessibility.  Most importantly, an effort was made to distinguish the 

distributional impact from the growth impact by controlling regional 

growth related variables.  In other words, the study attempted to control 

for the regional changes and then identify the "pure" distributional 

impact.  The results unveiled the fact that neither the existence of 

beltway itself nor the interchange density has consistently significant 

impact on population growth in SMSA during the study period (1960-1977).  

        Regarding the spatial analysis of land use change literature, a 

recent work, BART AT 20: A Preliminary Look at the Economic, Property 

Market, and Land Use Impact of the Bay Area Rapid Transit System, by John 

D. Landis (1994), studied the impact of Bay Area Rapid Transit stations 

on surrounding area land use conditions and evolution.  This study 

analyzed 1965-1990 land use changes at nine suburban BART stations.  

Grid-cell based analytical methods were used and probability models were 

developed in the study.  The results suggest that the distance to the 

existing BART stations have significant influences on land use change. 

The farther a grid-cell is away from BART stations, the less likely this 

grid-cell would change its land use.  In addition, adjacent land use type 

also stood out as a highly significant and consistently important 

variable, especially during the 1975-90 period.  The positive coefficient 

for this adjacent land use variable indicates that land use of each 

individual grid-cell is strongly affected by the land use pattern of its 

neighboring, or adjacent grid-cells (the existence of the spatial 

neighboring effect).    



Research Questions

        As mentioned before, the main purpose of this study is to 

demonstrate the impact of interstate highway systems on population 

decentralization since the 1960s.  In order to achieve this goal, the 

following three major research questions are proposed.

1. Spatial correlation: Are the expansions of urbanized areas spatially 

correlated with the development of interstate highway systems?  That is, 

do the directions of population decentralization match the directions of 

highway system development?  Figure 3 illustrates the spatial correlation 

effects.  The inner shaded circle stands for the existing urbanized area 

before the highway systems were constructed.  The thick black line 

represents a radial highway system which goes right through center city 

area.  It is assumed that the net expansion of urbanized area are equal 

for both types of growth (A and B).  Should there be no spatial 

correlation effect, type A expansion is expected.  Otherwise, the new 

urbanized area would look like type B.

2. Spatial neighboring effects: Do spatial neighboring effects play a 

significant role in the expansions of urbanized areas?  That is, all else 

being equal, do lands which are closer to the existing urbanized land 

have a higher chance of changing their urbanization status?

3. Time lag effects: Do time lag effects of highway systems on population 

suburbanization exist?  With other conditions controlled, would a 

non-urbanized land cell which is closer to a younger highway interchange 

have a higher probability of turning into an urbanized land cell than 

that one whose closest highway interchange is older?



Figure 3, The concept of Spatial Correlation Effects



4. Beltway effects: Do beltway systems have any effect on encouraging 

in-fill type of urbanized area expansion (type A in figure 3) rather than 

radial expansion along the highway corridors like type B in figure 2? 

5. Transitions of spatial effects on suburbanization: Whether two major 

spatial effects, spatial correlation and spatial neighboring effects, are 

experiencing transitions in terms of their impacts on suburbanization.



Methodology

Case Study Area Selection: Oklahoma City, Oklahoma

        In this study, Oklahoma City, Oklahoma, is selected for 

longitudinal analysis over three decades (1960-1990) of interstate 

highway evolution and urbanized area growth.  The following information, 

including some digital map files, about the case study area are collected 

for research purposes.

1. Urbanized area boundaries of case study area for year 1960, 1970, 

1980, and 1990

2. Street center line maps for the case study area (digital format)

3. Interstate highway system evolution from 1960 through 1990

4. Interchange locations and their years of completion

5. Locations of undevelopable lands, e.g., rivers (digital format)

6. Other arbitrarily defined boundaries, e.g., city boundary and county 

boundary (digital format)

7. Classification of highway types, i.e., beltway vs. radial systems.
        Figure 4 provides an overview of the evolution process of 

urbanized area of Oklahoma City from 1960 through 1990.  In this study, 

spatial models are developed to distinguish the directional effects of 

the interstate highway system from its density effects on urban 

decentralization.

GIS Analysis

        By using a GIS software (ARC/INFO for UNIX running on IBM RS6000 

platform), this research develops several grid-cell based spatial 

analysis models to explore the spatial evolution process of Oklahoma City 

urbanized area and its relationships with interstate highway 

constructions between every observation time period (every 10 years) or 

every other observation time period (every 20 years) .  In the following 

multivariate logit models, change of urbanization status (dichotomous 

variable) will be used as the dependent variable throughout all models.  

Each cell is treated as a research unit and each non-urbanized cell's 

probability of changing its urbanization status in the next decade is 

represented as a nonlinear function of its distance to the closest 

existing urbanized cell, its distance to the closest highway interchange, 

year of completion of the closest highway interchange, and type 

(beltway/radial) of the closest highway interchange.



Cell size selection

        The urbanized area boundary coverages for Oklahoma City, Oklahoma 

are determined from census data issued every 10 years by the Census 

Bureau.  Due to lack of precision of the hard copy maps in the census 

books, the source of the urbanized area data, cell size is set to equal 

one half square mile per cell (about 3734 feet on each side).  In other 

words, this study employs neighborhood-scaled observations in the models.
The definition of study areas

        Empirical evidence shows that the absolute magnitude of urbanized 

area growth is actually a function of existing urbanized area's size.  

Larger urbanized areas tend to have larger expansion.  Thus, a study area 

should be defined in a way which is sensitive to the area's size.  It is 

clear that traditional methods for defining study areas, such as using 

politically defined boundaries, are not appropriate.   Besides, the 

numbers of changed cells and non-changed cells should not be heavily 

weighted to either side in order to make the forecast model more 

meaningful.   Therefore, only those cells which are perceived by authors 

to have a reasonable chance of changing their urbanization status are 

included in the study area.  In this case study city, Oklahoma City, 

Oklahoma, the study area is defined as a 3-mile buffer of 1960 urbanized 

areas for 1960-1970 model, 3.5-mile buffer of 1970 urbanized area for 

1970-1980 model and 1960-1980 model, and 4-mile buffer of 1980 urbanized 

area for 1980-1990 model as well as 1970-1990 model .  Within each study 

area boundary, areas covered by water body, e.g., lakes and rivers, are 

regarded as undevelopable areas and thus are taken out of the sample 

set.
Logit Model Development

        Among others, there are three commonly used probability models: 

Linear Probability Model (LPM), Logit Model, and Probit Model.  The 

difficulty with LPM is its linear assumption which may result in 

observations with probability greater than 1 or less than 0, a 

meaningless result in a probability model.  Both logit model and probit 

model belong to the nonlinear probability model family with values always 

bounded by zero and one.  Usually, both models have similar performance 

in terms of their prediction power and accuracy rate.  However, the logit 

model is relatively easier to be interpreted.  In this study, the logit 

model with dichotomous dependent variable (change of urbanization status 

during the study period, 0 = no changed, 1 = changed) is used with each 

grid-cell in the defined study area acting as a unique observation.  The 

basic format for dichotomous logit model is as follows:

        P(Y = 1 | X) = exp( sum(bx))/[1 + exp( sum(bx))]

        where X are observed variables and the remaining unknowns are the 

parameters b.
Spatial correlation phenomenon

        It is commonly believed that the massive interstate highway 

system construction since the mid 1950s' is one of the major causes for 

the rapid growth of urbanized lands.  The Euclidean distance of each 

cell's centroid to its closest highway interchange is measured to 

represent the actual distance between the two.  This independent variable 

(DIST_INT) is expected to have significant negative coefficients 

throughout all multivariate logit models.
Spatial neighboring phenomenon

        By definition, the urbanized area is a cluster of land cells with 

their population density reaching a certain threshold.  It is assumed 

that the presence of an urbanized area cell can enhance the chance of 

urbanization of its neighboring cells.  Authors assume that the evolution 

of urbanized areas was affected by interstate highway systems but also by 

the spatial neighboring effect.  The rationale for expecting the 

neighboring effect is land developers' avoidance of costly extension of 

urban infrastructure.  Using the census definition of urbanized area as 

places with population densities greater than 1000 persons per square 

mile implies the need for water and sewer line infrastructures in the 

newly developed lands.  Land developers are inclined to develop land near 

existing systems to avoid costly mainline extensions.   

In order to distinguish the interstate highway spatial correlation effect 

from spatial neighboring effect, the latter is controlled by adding an 

independent variable, DIST_UA which measures the Euclidean distance from 

each observed cell to its closest urbanized area cell during the previous 

decade.  Consequently, significant and negative coefficients for this 

predictor (DIST_UA) are expected throughout all logit models.  That means 

the farther a cell is away from existing urbanized cells, the less likely 

this cell will turn into an urbanized cell.
Time lag effects

        There remains an unknown issue regarding whether a time lag 

effect exists in terms of encouraging non-urbanized areas growing along 

highway corridors.  Urbanized area's growth directions may not reflect 

the influences of nearby highway interchanges until much later than the 

interchange's completion year.  In order to explore this issue, this 

research includes the closest highway interchange years of completion as 

another independent variables (INT_60, INT_70, and INT_80).  These time 

lag variables appear as dummy variable formats on cell by cell base.
Beltway effects

        As mentioned, beltway systems are believed to play important 

roles in forming urbanized area boundaries.  If a non-urbanized cell's 

nearest highway interchange is a beltway interchange, it is expected to 

have higher probability of changing its urbanization status (positive 

effect).  In the following logit models, dummy variable (INT_BELT) is 

used to explore the beltway effect on urbanization.
Spatial Effects Transitional Analysis

        Based on logit modeling results, several curve fitting 

techniques, such as logistic curve fitting, are used to depict the 

historical trend and predict the future trend of spatial effects on 

suburbanization.  The probability of changing urbanization status during 

different time periods are then discussed.  The aggregate trend of 

urbanization area growth (without directional concerns) is used to adjust 

for the predicted results.  





Modeling Results

Logit Modeling

        In this study, a framework of the binomial logit model for 

predicting the probabilities of changing urbanization status among 

non-urbanized lands is specified as follows:
1960-1970 Model

        Figure 5 shows the urbanization area expansion between 1960-1970, 

highway systems as of 1960, and the study area (3 mile buffer of
1960  urbanized area boundary) defined for developing logit model.   For this  1960-1970 model, both DIST_INT and DIST_UA unveil significant explanatory  power on UA_CHANGE.  DIST_UA plays a more important role in predicting  the change of urbanized status than DIST_INT.  The negative coefficients  of DIST_INT and DIST_UA support previous hypothesis which suggest that  both the distance to nearest interchange and the distance to the nearest  urbanized cell have negative effects on the probability of changing  urbanization status for non-urbanized cells.  Nevertheless, this  1960-1970 model is not predicting any cell with urbanization status  change even though with 90.25% overall fit percentage.  My explanation is  that the sample set is highly skewed with 657:71 no-change/change ratio  and two distance variables can only explain a limited portion of the  dependent variable's variance although both of them are statistically  significant.   The beltway effect does not appear to be significant.  In  fact, based on the status of the Oklahoma City highway systems as of  1960, as shown in figure 5, no beltway effect should have been expected. 

Table 1, Logit Modeling Results: 10-Year Models 
Dependent Variable      UA_CHANGE                                         Study Period            1960-1970               1970-1980               1980-1990 
DIST_INT                -0.1757( .0000)         -0.0519( .1707)         -0.1834( .0000) DIST_UA                 -0.6911( .0000)         -1.1726( .0000)         -0.6823( .0000) INT_60                                          0.0803( .6784)          -0.5089( .0343) INT_70                                                                  -0.1093( .5499) INT_BELT                -0.1305(.6824)          0.0162(.9619)           3.4991(.0010) 
Constant                .1853( .5875)           0.5575( .0085)          1.0476( .0000) 
Observations            728                     908                     1075 Change                  71                      208                     327      Predicted          0                       0                       95      % Predicted        0%                      0%                      29.05% No Change               657                     700                     748      Predicted          657                     700                     696      % Predicted        100%                    100%                    93.05% Overall Fit             90.25%                  77.09%                  73.58% 
Note: Numbers in the parenthesis show the t statistics for the  corresponding variables.  Numbers with bold font are statistically  significant at 0.05 level. 

1970-1980 Model         Figure 6 shows the urbanized area expansion between 1970-1980,  highway systems as of 1970, and the study area defined for developing  logit model.  A total of 908 cells are covered by 3.5-mile buffer of 1970  urbanized area.  Similarly, in the second, 1970-1980 model, the signs of  coefficients of both distance variables, DIST_INT and DIST_UA, are  negative as expected (-0.0519 and -1.1726 respectively).  The coefficient  of DIST_UA is once again larger than the coefficient of DIST_INT which  indicates that the distance to the nearest urbanized area cell (spatial  neighboring effect) played a more important role in predicting urbanized  area expansion during 1970-1980.  In fact, the coefficient for DIST_INT  appears to be insignificant with its P value equals to 0.1707.  The  positive coefficient (0.0803) of INT_60 (time lag effect dummy variable)  suggests that a cell would have a better chance to become an urbanized  cell between 1970 to 1980 if its closest highway interchange was  completed by 1960 (rather than 1970).  However, this coefficient is not  statistically significant (p = 0.6784).  The beltway effect once again  does not appear to be significant (b = 0.0162, p = 0.9619).  Same as the  1960-1970 model, this 1970-1980 model is not predicting any cell with  urbanization status change even though with 77.09% overall fit  percentage.  The same explanation as mentioned in the previous 1960-1970  section can also be applied here. 1980-1990 Model         Figure 7 shows the urbanized area expansion between 1980-1990,  highway systems completed by 1980, and the defined 4-mile buffer (of 1980  urbanized area) study area.  The total number of observed cells equals to  1075.  The signs of coefficients for both distance variables are negative  and significant as expected.  Consistently, the distance to the nearest  urbanized areas continues to act as a more influential factor than the  distance to the nearest highway interchange.  The strong significance  level (relatively high t statistics) of both coefficients reflects that  both distance variables are justified to have their own effects on the  change of urbanization status.  The negative signs of time lag variables,  INT_60 and INT_70, shows that the newer interchanges completed between  1970 and 1980 are more likely to be associated with new development.   Meanwhile, the coefficients for both time lag effect dummy variables,  INT_60 and INT_70, are not proven to be statistically significant.   Surprisingly enough, unlike previous two decades, the beltway effect  comes up significantly (p = 0.001) with a relatively large coefficient  (3.4991).  The beltway effect has become as a strong factor along with  others that influence urban sprawl directions in the 1980s.  Twenty-nine  percent (29%) accuracy of predicting urbanized cells is achieved by this  model compare to 0% in previous models.
Limitations of 10-year Models and the Concepts of 20-year Models

        Because of the inherent constraint of our arbitrarily defined 

study areas, e.g., 3-mile buffer, 4-mile buffer, etc., it would be unfair 

to conclude that the widely held expectation that highways encourage 

urban sprawl has proven to be incorrect.   For example, figure 5 shows 

the change of urbanized area boundaries between 1960 and 1970.  It is 

obvious that south portion of urbanized area growth is well beyond the 

3-mile buffer of 1960 boundary.  In order to avoid the potential of 

losing information, we developed another set of models which examine the 

urbanized status change in every 20 years.  That is, two models, one for 

year 1960 to 1980, the other for year 1970 to 1990, are developed based 

on a very similar framework.  For the 1960 - 1980 model, the distance to 

the nearest highway interchange is recorded for each cell based on 

highway networks completed by 1970.  The distance to the nearest existing 

urbanized area cell is measured based on 1960 urbanized area boundary.  

Correspondingly, for the 1970-1990 model, the distance to the nearest 

highway interchange is based on highway networks by 1980 while the 

distance to the nearest urbanized cell is measured based on 1970 

urbanized area boundary.  Table 2 summarizes the results of both models.
1960-1980 Model

        Figure 8 shows the urbanized area expansion between 1960-1980, 

highway systems completed by 1970, and the defined 3.5-mile buffer (of 

1970 urbanized area) study area.  For 1960-1980 model,  both distance 

variables coefficients show significant explanatory power.  Unlikely 

previous 10 year period models, the distance to the nearest highway 

interchange variable (b = -0.1881) becomes a more influential factor in 

changing urbanization status than the distances to the nearest existing 

urbanized cell variable (b = -0.0752).  Once again, the time lag 

variables, INT_60, is not a significant factor in this model.  As 

expected, the beltway effect, INT_BELT (b = 0.2775), is not a significant 

predictor.  The accuracy rate of prediction is 68.32%.  Apparently, this 

model is much more reliable when predicting a cell as no change (0 to 0, 

96.71% accuracy rate) than it is when predicting a cell as change (1 to 

1, 11.68% accuracy rate).  However, this is already an improvement in 

contrast to the 1960-1970 model and the 1970-1980 model where cells with 

changed urbanization status are predicted with 0% accuracy rate.  










Table 2, Logit Modeling Results: 20-Year Models









Dependent Variable      UA_CHANGE                         Study Period            1960-1980                       1970-1990
DIST_INT                -0.1881( .0000)                 -0.1486( .0000) DIST_UA                 -0.0752( .0044)                 -0.4223( .0000) INT_60                  0.1015( .5195)                  -0.4539( .0267) INT_70                                                  -0.2396( .1333) INT_BELT                0.2775(.3468)                   3.3636(.0011)
Constant                0.1672( .2704)                  1.3990( .0000)
Observations            1051                            1299 Change                  351                             551      Predicted          41                              313      % Predicted        11.68%                          56.81% No Change               700                             748      Predicted          677                             597      % Predicted        96.71%                          79.81% Overall Fit             68.32%                          70.05% 
Note: Numbers in the parenthesis show the t statistics for the  corresponding variables.  Numbers with bold font are statistically  significant at 0.05 level.

1970-1990 Model         Figure 9 shows the urbanized area expansion between 1970-1990,  highway systems completed by 1980, and the defined 4-mile buffer (of 1980  urbanized area) study area.  In this 1970-1990 model, both distance  variables appear significantly with negative coefficients as expected.  The distance to the nearest urbanized area cell, DIST_UA (b = -0.4223) is  a more important factor in terms of its impact on urbanization status  change than DIST_INT (b = -0.1486).  Table 3 compares the coefficients of  two distance variables derived from both 1960-1980 model and 1970-1990  model.  On one hand, the decrease of coefficient value of DIST_INT  indicates that the impact of highway systems on suburbanization is  declining.  On the other hand, the increase of coefficient value of  DIST_UA suggests that the spatial  neighboring effects on suburbanization is growing.  A further discussion  regarding this phenomenon will be made in the concluding section.  The  coefficient of INT_BELT unveils the same information as previous  1980-1990 model that beltway effect has tremendous predict power (b =  3.3636, p = 0.0011) during the last 10 to 20 years or so.  The negative  and significant coefficient of INT_60 suggests that having a more than 20  years old highway interchange as the nearest interchange, a non-urbanized  land cell would have less chance of changing its urbanization status than  others.

Table 3, Comparison of Distance Variables' Coefficients between 20-year models
Variable                1960-1980 model         1970-1990 model         % change
DIST_INT                -0.1881                 -0.1486                 - 21%
DIST_UA                 -0.0752                 -0.4223                 + 462%
Stronger Predictor*     DIST_INT                DIST_UA 
* Predictor's strength is decided by its absolute value of coefficient



Interpretation of Models' Coefficients
        Because logit model is a nonlinear probability model, the
influence of each predictor is not constant.  In other words, the
interpretations of coefficients are not very straight forward.  Using
1960-1980 model as an example, each mile increase in the distance to the
nearest highway interchange is estimated to decrease the odds of changing
from non-urban to urbanized cell by a factor of 0.829 ( = exp(-0.1881)),
or about 100 ( 1-0.829) = 17.1 %.  Similarly, each mile increase in
distance to the nearest urbanized cell is estimated to reduce the odds of
changing from non-urbanized to an urbanized cell by a factor of 0.924 ( =
exp(-0.0792)).  Coefficients in the 1970-1990 model can be interpreted
accordingly.  For example, one is examining a non-urbanized cell in 1970
which is 1.25 miles away from existing urbanized cell, 2 miles from the
nearest interchange.  Its nearest highway interchange was completed by
1970.  The value of odds is e (0.1979 - 0.19008 * 2 - 0.0792 * 1.25) = e
-0.28126 = 0.7548.  For the case of logit model with dichotomous
dependent variable, P / (1 - P ) is equal to 0.7548, and therefore P
equals to 0.4301 < 0.5.  The predicted value for this land cell would
equal to 0 (no change).  Correspondingly, in 1970-1990 model, a cell with
identical conditions (1.25 miles to urbanized cell and 2 miles to
interchange) would have odds equal to exp (1.4357 - 0.13728 * 2 - 0.45936
* 1.25 - 0.2167) = 1.4481.  P is then equal to 0.5915 (P / ( 1 - P ) =
1.4481) which is greater than 0.5.  The predicted value for this land
cell would be 1 (change).

Spatial Effects Transitional Analysis
        Based on previous modeling results, I conclude that the spatial
neighboring effect is playing a more important role than the spatial
correlation effect, which was a dominant factor during the massive
highway construction era.  In order to explore this phenomena more
clearly, I use curve fitting techniques to depict this transition
graphically, as well as to predict the future urbanization probabilities
for non-urban lands with different physical properties, e.g., distance to
interchange.  Twenty-year modeling results are used as the base of this
transition analysis.
        Figures 10 and 11 illustrate the probabilities of urbanization of
lands with various distances to the existing urbanized area and the
nearest interchange.  As indicated, the impact of the two spatial effects
can be examined through the differences of slopes.  A larger difference
between the slope of the axis and the slope of observed probabilities,
when controlling for the other axis (effect), reveals a greater impact of
that spatial effect.  We can see in Figure 10 that the slope facing the Y
axis (spatial correlation effect) is steeper than that facing the X axis
(spatial neighboring effect).  Once again, a stronger spatial correlation
effect for the 1960-1980 period is confirmed.  In Figure 11, the opposite
result is observed.  The slope facing the X axis is much steeper than
that facing the Y axis in the 1970-1990 period.  Clearly, a spatial
effect transition took place between the two time periods.  If there are
no other major technological and policy changes in the near future, the
spatial neighboring effect is expected to play a more important role in
population suburbanization in the next few years.
        Figure 12 shows the probabilities of changing urbanization status
for lands 1) five miles from nearest interchange and one mile away from
existing urbanized area (UA1INT5, solid line in the Figure), or 2) one
mile from nearest interchange and five miles away from existing urbanized
area.  Both types are controlled for the age of the nearest interchange
as "1970" and the type of the nearest interchange variables as
"beltway".  Time period 0 reflects the probabilities during the 1960-1980
period while time period 1 stands for 1970-1990 period.  In other words,
curve segments beyond time period 1 are the predicted probabilities.  The
logistics curve fitting model is used due to its desirable bounded growth
property (probability * 0 and * 1).  As expected, distance to the
existing urbanized area is becoming more and more decisive in terms of
influencing the urbanization probabilities.  Any developable land which
is close to the existing urbanized lands, e.g., UA1INT5, has a tremendous
potential to become urbanized land in the near future.
 

Figure 12, Transitional Trend of Probability of Urbanization
Note: UA1INT5 stands for lands with one mile away from existing urbanized
area and five miles from the nearest interchange.  UA5INT1 stands for
land with five miles away from existing urbanized area and one mile from
the nearest interchange.

Conclusions

Policy Implications


        Intelligent Transportation Systems (ITS) is considered by some as 

a major contribution to transportation technology as we move toward year 

2000.  The deployment of ITS technologies is expected to influence 

people's travel behavior in the near future.  Among various ITS goals, 

saving travel time, both individual time and commercial time, is one of 

the most important components.  Many time-saving related technologies, 

such as adaptive traffic signal control systems and dynamic route 

guidance systems, are either commercially available or under field test 

status.  From the lessons of interstate highway development in the past 

forty years, we have already learned that people show their willingness 

to trade travel time savings for greater travel distance to meet their 

desires for low density neighborhoods and low density workplaces.  Urban 

sprawl in the past decades has indeed caused enormous social and 

environmental costs.  We have made mistakes and have paid for them.  We 

want to make sure that ITS can be deployed in such a way that people are 

encouraged to preserve the benefits of travel time saving rather than 

substitute them with any other potentially detrimental decisions, such as 

moving further away.  While we are lucky enough to see that the driving 

force for suburbanization is getting weaker, we have to prevent other 

causes from emerging.  The potentials of some ITS technologies, such as 

Automated Traffic Management Systems (ATMS) and Automated Traveler 

Information Systems (ATIS) for encouraging urban sprawl should be 

properly addressed.  And the benefit cost analysis of ITS deployment 

should also be conducted with extreme caution so as not to overestimate 

ITS benefits, e.g., travel time saving, with the assumption of current 

travel patterns remaining unchanged.

        From experiences, our society has learned that a net increase in 

people's mobility tends to enhance the magnitude of urban sprawl, i.e., 

the amount of urbanized area expansion.  An unbalanced distribution of 

social resources, e.g., income and mobility, will contribute to the 

sprawl of urban forms.  The combination of rapid and unbalanced urban 

area expansion causes enormous problems, such as traffic congestion and 

some derived environmental problems, such as air pollution.  Many 

transportation-land use studies in the past explored many different 

solutions, such as job-house balancing strategy, to cure the urban sprawl 

problems.  According to Levine (1995), one extreme of thought argues that 

we should let the market mechanism work by itself.  In contrast, the 

other extreme of thought, in contrast, proposes that a stricter land use 

control procedure should be enforced.  Both ideas have proven to be  

inefficient solutions.  If the market could work by itself, we would not 

have had this urban sprawl problem in the first place.  There exist 

negative externalities associated with the residential location decisions 

which create the improperly distribution of social resources.  We saw 

richer people moving to the suburbs, with the social costs of their 

relocating behaviors burdening the whole society.  This distorted market 

then creates some inefficient results.  On the other hand, local land use 

control seemed to serve richer people better than it did to the poor.  In 

many cases, we saw clearly that richer people moved to the suburbs and 

then used land use control as an excuse to keep poor people from 

following.  Even the job-house balancing policy does not appear to be an 

effective solution to this sprawl problem for several reasons, such as 

the fact that many household have two or more workers going in various 

directions. 

        Considering the fact that the highway effect on suburbanization 

is diminishing and ITS technologies are emerging, I am proposing the 

following four policy suggestions to cope with the current transportation 

problems and related issues.
1. Internalize suburbanization externalities

        People who move to suburban areas and generate negative 

externalities should be paying for the "true social cost" of their 

choice.  The higher cost would have the effect of discouraging urban sprawl.

2. Re-prioritize ITS deployment procedures

        Many social policies, such as taxation, aim at income 

redistribution.  In many respects, mobility can be treated as another 

type of property owned by individuals.  Most ITS development efforts in 

the U.S. seem to focus on the private and commercial transportation 

services.  It is important that public transportation applications of ITS 

are also promoted in such a way that public transit users are not 

discriminated against by this technology.  This is very crucial in the 

sense that unbalanced distribution of mobility is also a major cause of 

urban sprawl.  The potentials of ITS for private transportation are 

expected to enhance the users' mobility significantly.  The deployment 

process of ITS user services should be scheduled in such a way that 

public transit users (usually having less mobility) can receive the 

benefits of ITS at the same time as private mode users, if not earlier.

3. Equip people with resources and provide them with more options

        The major reason why job-house balancing did not work well is 

that no one knows how to meet people's needs better than people 

themselves.  Instead of offering people the "assumed" best choice, people 

should be given the power to make their own choices.  In this case, 

people should be allowed to live any place they want (of course, with 

certain restrictions), as long as they are willing to pay the "true 

cost".  Government's role is to provide people with similar resources, 

e.g., income, education, etc. so that they will have adequate knowledge 

and information to maximize their personal utilities.

4. Let market run by itself when the situation "matures"

        In most cases, government interventions became the biggest cause 

of inefficiency and society loss.  I agree that to a certain extent 

government intervention is necessary.  But the goal of government 

intervention should aim at maintaining a healthy environment and social 

justice (i.e., prevent the more powerful from ripping off weaker people), 

instead of providing people with "end products" that government believes 

are best.  In the ITS-transportation case, the government should make 

sure of the existence of equal opportunities in terms of selecting house 

and workplace locations.  Then, let the market mechanism generate the 

optimal solution. 

        It has been 38 years since the passage of Federal-Aid Highway Act 

of 1956.  The passage of Intermodal Surface Transportation Efficiency Act 

(ISTEA) of 1991 authorized $21 billion out of its $151 billion total 

budget over six years for national highway system construction and 

maintenance.  As a response to the ISTEA, the Secretary of 

Transportation, Federico Pena, brought the proposed 158,000-mile National 

Highway System (NHS) before Congress in December 1993, giving birth to 

the National Highway System Designation Act of 1994.  In other words, 

tremendous attention to and action on the interstate highway system is 

foreseeable.  Being urban planners, we have observed the profound impact 

of highway systems on urban forms.  We have recognized the problems with 

certain types of urban sprawl.  Hopefully, this research will contribute 

to clarifying factors influencing social benefits of highway investments.



Future Study Suggestions

Expand the Scope of this Study

        This research can be expanded on three different dimensions: 

issue, space, and time.  On the first dimension - issue, I would like to 

explore the transitional dynamics between interstate highway systems and 

other social phenomenon, such as population, migration, and environment 

dynamics.  On the second dimension - time, I intend to include a longer 

period of time, e.g., starting from the pre-auto era to the present.  

Finally, on the third dimension - space, I wish to extend the study to 

examine more cities in the U.S. or even the rest of the world, in order 

to categorize transitional patterns.  This study should be treated as a 

pilot study, to be followed by a series of  further in-depth 

explorations. The final goal of this study is to develop a comprehensive 

methodology which is replicable to other urbanized areas in/out of the 

U.S.  



Accommodate Multi-Level Model Structure

        Apparently, the proposed model would encounter the dilemmas of a 

hierarchical data structure.  The Level-1 units are grid-cells, which are 

nested within the Level-2 units of highway completion year.  As indicated 

by A. S. Bryk and S. W. Raudenbush (1992), the most common concerns 

regarding hierarchical data analysis include aggregation bias, 

misestimated precision, and the "unit of analysis" problem.  

Nevertheless, this study would still use a conventional dichotomous logit 

model to predict the probability of urbanized status change due to the 

following two reasons:
Add Zoning/Land Use Information to Exclude those Undevelopable Lands

        In this study, only water body is excluded from the sample set.  

In the real world, many types of lands are not developable due to 

limitations, such as unsuitable soil type, and environmentally sensitive 

and preserved areas.  The forecast accuracy rate would be improved, if, 

before establishing models, future studies excluded these non-qualified 

areas from the sample set.

Use Discrete Network Distance to Measure Real Travel Time rather than 

Making Continuous Space Assumption

        This study uses a simplified distance measurement method, 

Euclidean distance, to acquire distance information.  An effort to apply 

the discrete network concept to measure actual travel time hopefully 

would improve the models' prediction capability.

Conduct Residual Autocorrelation Analysis for each Prediction Model

        This study relies on the "eyeballing" method to examine modeling 

residuals' spatial autocorrelation phenomena.  In fact, there are several 

spatial autocorrelation models available to explore the intensity of 

autocorrelation in a quantitative manner, e.g., Moran's I statistics and 

joint-count statistics (Upton and Fingleton, 1985).  It is desirable to 

apply these available analytical tools to justify the models' assumptions 

and to improve their 

Definition of Beltways

        The most commonly used method to define and label a beltway is to 

base it on the highway numbers, such as found in Payne-Maxie Consultants, 

1980.  For example, I-96 is a main line while I-696 is a beltway.  The 

same is true for I-75 vs. I-275.  In this study, I also used this 

intuitive way to define beltway for the case study area.  Nevertheless, I 

recognize that a set of objective and scientific criteria are necessary 

for academic research purposes, especially for a geographic analysis like 

this.  I am currently developing a method of defining beltway from a 

geometric perspective.  Hopefully, the follow-up study can apply this new 

method to objectively defined beltways.
REFERENCES
1. John H. Aldrich and Forrest D. Nelson, Linear Probability, Logit, and 

Probit Models, SAGE publications, Newbury Park, California, 1984.
2. Katharine Bradbury, Anthony Downs, and Kenneth Small, Urban Decline 

and The Future of American Cities, The Brookings Institution, Washington 

D.C., 1982.
3. Anthony S. Bryk, Stephen W. Raudenbush, Hierarchical Linear Models: 

Application and Data Analysis Methods, Sage Publications, 1992.
4. Keith C. Clarke, Analytical and Computer Cartography, Prentice-Hall, 

Inc., New Jersey, 1990.
5. Judy S. Davis, Arthur C. Nelson, and Kenneth J. Dueker, "The New 

Burbs,: The Exurbs and Their Implications for Planning Policy", JAPA, 

Winter 1994, pp. 45-60.
6. Alfred Demaris, Logit Modeling: Practical Applications, SAGE 

publications, Newbury Park, California, 1992.
7. Anthony Downs, Opening Up the Suburbs, Yale University Press, New 

Heaven and London, 1974.
8. Anthony Downs, Stuck in Traffic, The Brookings Institution, Washington 

D.C., 1992.
9. Eran Feitelson, "The Spatial Effects of Land Use Regulations: A 

Missing Link in Growth Control Evaluation", JAPA, Autumn 1993, pp. 461-472.
10. Susan Hanson, The Geography of Urban Transportation, The Guilford 

Press, New York, 1986.
11. Jonathan C. Levine, "Decentralization of Jobs and the Emerging 

Suburban Commute", Research Report, College of Architecture and Urban 

Planning, The University of Michigan, 1992.
12. Mitchelle E. Macdonald, "Planned National Highway System to Upgrade 

158,000 Miles of Roads", Traffic Management, February 1994, pp. 17-18.
13. Payne-Maxie Consultants, The Land Use and Urban Development Impacts 

of Beltways, Final Report No. DOT-OS-90079, U.S. Department of 

Transportation and Department of Housing and Urban Development, 

Washington, D.C., 1980.
14. Eudora S. Pendergrast, Suburbanizing the Central City : An Analysis 

of the Shifting in Transportation Policies Governing the Development of 

Metropolitan Toronto, 1959-1978, Department of Urban and Regional 

Planning, University of Toronto, 1981.
15. F. K. Plous, Jr., "Refreshing ISTEA", Planning American Planning 

Association, February 1993, pp. 11-16.
16. Michael Southworth and Peter Owens, "The Evolving Metropolis: Studies 

of Community Neighborhood, and Street Form at the Urban Edge", JAPA, 

Summer 1993, pp. 288-305.
17. Jon C. Teaford, The Twentieth-Century American City: Problem, 

Promise, and Reality, The Johns Hopkins University Press, 1986.
18. Graham J. G. Upton and Bernard Fingleton, Spatial Data Analysis by 

Example, John Wiley & Sons, 1985.
19. Sammis B. White, Lisa S. Binkely, and Jeffrey D. Osterman, "The 

Sources of Suburban Employment Growth", JAPA, Spring 1993, pp. 193-204.
NOTES